Riot Headline Book Riot’s 2025 Read Harder Challenge
Image of a book on fire with the words "book censorship news" in white beneath the image
Censorship

Districts Are Turning to AI to Ban Books: Book Censorship News, August 18, 2023

Kelly Jensen

Editor

Kelly is a former librarian and a long-time blogger at STACKED. She's the editor/author of (DON'T) CALL ME CRAZY: 33 VOICES START THE CONVERSATION ABOUT MENTAL HEALTH and the editor/author of HERE WE ARE: FEMINISM FOR THE REAL WORLD. Her next book, BODY TALK, will publish in Fall 2020. Follow her on Instagram @heykellyjensen.

View All posts by Kelly Jensen

Across several U.S. states, schools are scrambling to figure out how they can be in compliance with new laws implemented over the summer which change the kinds of materials allowed in their classrooms and libraries. Iowa is one such state. Senate File 496 requires all materials be age appropriate and that there be no “descriptions or depictions of sex acts,” defined by Iowa Code 702.17. The state has given virtually no guidance beyond what’s written, and it is a law championed and celebrated by several chapters of Hate Group Moms for Liberty throughout the state. It is an impossible task without a time limit, and it is made even more impossible when the state gives no direction and expects districts to be in compliance within weeks. How do schools read every book in the collection and make a decision whether or not it is age appropriate or contains “sex acts?”

Mason City School District (MCSD) found themselves in this very position and turned to an unexpected source to determine the status of books in their collection: AI. With the use of artificial intelligence, the school district found 19 books to be out of compliance with the new law and removed the titles.

Curious about the AI tool they used, I reached out to the district and got an answer. They used ChatGPT.

Before explaining how they used the tool to determine appropriateness and whether or not it holds up on retest, it should be noted that in the last 20 years, the district has received zero challenges from parents for their collection. They have a policy for filing challenges, so they’ve been practicing their due diligence since the start — this is how they determined that they were in compliance with the first part of the new law.

It is the use of AI that raises significant concerns. Again: the district used this tool because they are under time constraints put upon them from the governor’s office. Their decision to do this should not indicate maleficence on their part but on the part of Governor Kim Reynolds and every member of the legislature influenced by the false rhetoric and money tied to Moms for Liberty and similar groups. The use of ChatGPT is done out of desperation and this is what should be of concern, especially as the district has within their student handbook that plagiarism — “copying from other sources” — is considered cheating. The district does not specify where or how AI plays into their definition of academic dishonesty.

To determine whether or not materials were in compliance with the new law, the district did the following: they found lists of books on the internet which had been challenged and compiled a master list. They then removed titles on the master list which were not challenged based on “sex acts.” The third step in their process was to search their collection for those titles, winnowing down the list to just the most commonly challenged books that were their libraries.

The final step? Running each of those titles through ChatGPT with the question “Does BOOK NAME contain a description or depiction of a sex act?” If the answer to the query was yes, the book was removed (and, “fortunately,” placed in storage, rather than discarded permanently, though this still falls squarely under the definition of banning a book). MCSD is smart to keep their query to the exact definition of the new state law.

Knowing that ChatGPT is trained through what’s already out there on the internet, no doubt several right-wing, book banning websites were key in its knowledge. Websites from other sources were, too, but those sources will inevitable use the phrase “sex” or “sex acts” or some variation therein to describe why the book was challenged. As such, the answer to any query in ChatGPT — being that it is AI and does not consider context in the way actual human intelligence does — would be likely to bring that up. But it is AI, so there are times when it is spot on (and frankly far more open minded than right-wing book banners are right now: see this!).

So how does this look in real time?

First, note the warning that pops up in using such a query in ChatGPT. It may violate the tool’s content policy. Let’s start with one that very certainly does not contain “sex acts,” but which has still seen pushback in some municipalities: Romeo and Juliet. I used the exact query used by MCSD.

Image of Chat GPT query does ROMEO AND JULIET contain a description or depiction of a sex act This content may violate our content policy. If you believe this to be in error, please submit your feedback — your input will aid our research in this area. ChatGPT No, "Romeo and Juliet" by William Shakespeare does not contain a description or depiction of a sex act. While the play does explore themes of love and passion between the titular characters, it does not provide explicit descriptions of sexual acts. The focus of the play is more on the emotional and tragic aspects of their relationship, as well as the societal conflicts that drive the story forward. It's important to note that Shakespearean plays typically use language that can be interpreted metaphorically, and any sexual content would likely be implied rather than explicitly described.

Then I asked ChatGPT to answer the same query with some of the 19 titles pulled by the district, starting with Jodi Picoult’s 19 Minutes.

Image of chat GPT query: Does "Nineteen Minutes" by Jodi Picoult contain a description or depiction of a sex act This content may violate our content policy. If you believe this to be in error, please submit your feedback — your input will aid our research in this area. ChatGPT "Nineteen Minutes" is a novel by Jodi Picoult that explores the aftermath of a high school shooting and its impact on the community. While the book addresses mature themes and includes discussions about relationships and sexuality, it is not primarily focused on explicit descriptions or depictions of sex acts. However, it's possible that the book may contain some references to sexual themes, but they are typically handled in a sensitive and context-appropriate manner. If you're considering reading the book and are concerned about its content, I recommend checking reviews or summaries to get a better understanding of the themes and content it addresses. Keep in mind that interpretations of content can vary from person to person, so what one reader might find explicit, another might not.

Well, ChatGPT doesn’t say it doesn’t contain “sex acts,” but does emphasize that those are in no way the primary purpose of the text.

The book was pulled anyway.

It is very likely that decision was made to follow in the conservative interpretation of the law. But it is interesting that ChatGPT offered a lot more insight into the content of this one than, say, Romeo and Juliet. This is probably because there’s simply more information about the “content” in Picoult’s book, thanks to it being a frequent favorite of the book crisis actors.

Now onto a book about a child sold into prostitution and how traumatic, damaging, and inhumane that is.

(Recall context doesn’t matter — we’re going only for whether or not the book contains “sex acts,” as defined by the Iowa Code).

Chat GPT query Does "Sold" by Patricia McCormick contain a description or depiction of a sex act? ChatGPT "Sold" by Patricia McCormick is a novel that addresses the serious and sensitive issue of human trafficking and the exploitation of young girls in the context of the sex trade. While the book does explore the experiences of the main character, a young girl named Lakshmi, who is forced into prostitution, it does so in a way that is appropriate for its target audience (young adult readers) and is focused on raising awareness about the issue rather than providing explicit descriptions or depictions of sex acts. The book handles the topic with sensitivity and is more concerned with the emotional and psychological impact of the situation on the characters. It aims to shed light on the harsh realities of human trafficking without resorting to graphic or explicit content. However, keep in mind that individual perceptions of what is explicit can vary, so if you are concerned about the content, I recommend reading reviews or summaries to make an informed decision about whether the book is suitable for you or the intended reader.

Again, ChatGT does not say it doesn’t, so the book was pulled.

The remaining 17 books pulled all returned similar responses from ChatGPT: there might be sexual moments or discussions in the text, but they are in no way the purpose of the story. If anything, ChatGPT surprised me with how much context it brought back on each title.

And yet, it’s now a handy tool to decontextualize texts further.

This is not the district’s fault. This is the district’s reaction to an impossible situation, and we’re going to see this mirrored over and over across the state and the country. Non-professionals in literacy and literature will ensure district compliance with outrageous laws via tools meant to make their lives easier, simply to be in compliance. I have no doubt for those in MCSD, these decisions were far from easy nor fun, but without the time, money, or staff to read through every single book and put it through a fair process involving reading the material, discussing it, and evaluating it. Moreover, other districts will be looking to this method in the weeks and months to come, replicating it so they, too, can continue being funded by the state.

But is using a tool that explicitly notes that using it that way isn’t okay even ethical? And what should we think about the fact that this isn’t doing your own work but is indeed skirting close to plagiarizing the work of others — and work that there is no references for to check on? Part of what students learn in classrooms and in school libraries is how to be information literature, including how to verify sources, determine their reliability, their biases, and their reputation. ChatGPT strips away all of this.

Indeed, using it to determine whether or not to remove books and stripping the little context offered by the AI tool only further devalues the work of professionals across the country aiming to create a literature, educated student body. If the schools don’t need to do all of the work, how or why do they expect that of the very kids they’re tasked to educate?

It’s because state does not want people who can think.

Makes sense, since they didn’t even think this law up themselves. They simply adapted it from the local hate groups and took them at their word.

Book Censorship News: August 18, 2023